Realized Genomic Selection Prediction Accuracies for End-Use Quality in CIMMYT Spring Wheat Breeding Program

Wheat cultivars must possess suitable end-use quality for release and acceptability. However, breeding for end-use quality traits is often considered a secondary goal compared to enhancing yield, disease resistance and stress tolerance, mainly due to limitations in amount of seed available in earlier stages, thus allowing many undesirable materials to be advanced. Our aim is to develop and validate genomic selection (GS) prediction models using end-use quality phenotypes routinely generated by the CIMMYT wheat breeding program on advanced wheat materials. Once model accuracy is determined, genomic selection can be applied at earlier stages and undesirable materials culled before implementing expensive yield and quality screenings. GS model accuracy was tested using data from CIMMYT first-year yield trials from 2009 to 2014 (n = 5,004 entries) grown for one year at Ciudad Obregon, Mexico, under high yielding environments and evaluated for grain yield and quality parameters, viz. test weight, grain hardness, grain and flour protein, mixograph dough mixing time, alveograph dough strength (W) and tenacity/extensibility ratio (P/L), and bread loaf volume. High-density markers were generated with genotyping-by-sequencing, and SNPs were imputed (n=32,180 SNPs). GS prediction was conducted with ridge regression from R package rrBLUP. 2009 – 2013 wheat lines (n=3,659) trained the model and predictions were made for 2014 entries (n=1,345) before phenotypic data was available. Prediction accuracies (r) for all quality parameters ranged from 0.417 (test weight) to 0.652 (mix-time). Based on these prediction accuracies, we conclude that GS can be a useful tool to facilitate early generation selection for end-use quality.